Being married reduces the mortality risk of older persons. More generally, living arrangements that include co-residence with a source of support and a close care-giver are associated with a lower mortality risk. We build a detailed typology of private and collective living arrangements, including marital status, and check its association with mortality risks, controlling for health status. Using administrative data from the population register, we identify the living arrangement of all individuals aged 65 years and over living in Belgium as at 1 January 2002, and their survival during the year 2002. Data on health status are extracted from the 2001 census. We use binary logistic regression with the probability to die as outcome and living arrangement, health, age and gender as covariates. Our results show that mortality is more closely associated with actual living arrangements than with marital status. This association is age and gender-specific and remains even at very old ages. Living with a spouse is confirmed to be beneficial for survival but in older age living alone becomes more favourable. Of all living arrangements, older persons living in religious communities experience the lowest mortality risk whereas those living in nursing homes experience the highest risk.